Drop top 1% and 10% of a variable

[Ronan Mack]

hello, I am wondering how to go about dropping the top 1% and 10% according to size of a certain variable. It is to be used as a robustness test for my data. Thanks

[Daniel Feenberg]

sort var
gen sum=sum(1)
keep if sum>.99

but see https://www.nber.org/stata/efficient/percentiles.html for some alternatives. The only documentation of sum() I can find at the moment is in https://www.stata.com/manuals/fn.pdf but it is pretty sketchy.

[Nick Cox]

The code in #2 will remove all your data. The new variable is by construction integers 1 up and the inequality is the wrong way round too. Further, the code is not subtle about missing values.

I believe that official command cumul offers what Daniel
is thinking of here. But so does summarize.

[Daniel Feenberg]

Of course I was too hasty and got it wrong. Should be (hopefully)

sort var
gen sum=sum(1-missing(var))
keep if sum>.99*sum[_N]

but as Nick points out, -cumul- is one step and returns the data to the original sort. It is documented at page 433 (out of 3077!) in https://www.stata.com/manuals/r.pdf

[Joro Kolev]

I would do it with -summarize-. Something like:

summ size, detail
keep if size< r(p99)

to drop the top 1% and

summ size, detail
keep if size< r(p90)

to drop the top 10%.